A Novel Deep Learning Approach for Anomaly Detection of Time Series Data

Anomalies in time series, also called “discord,” are the abnormal subsequences. The occurrence of anomalies series may indicate that some faults or disease will occur soon. Therefore, development novel computational approaches for anomaly detection (discord search) is great significance state monitoring and early warning real-time system. Previous studies show many algorithms were successfully developed used classification, e.g., health monitoring, traffic detection, intrusion detection. However, was not well studied. In this paper, we proposed a long short-term memory- (LSTM-) based method (LSTMAD) discord search from univariate data. LSTMAD learns structural features normal (nonanomalous) training data then performs via statistical strategy on prediction error observed our experimental evaluation using public ECG datasets real-world datasets, detects more accurately than other existing comparison.